如何在TensorFlow中获得None的真实形状（动态输入形状）？

Question

I have a placeholder which shape is [None, dimension], "None" means batch size. I want to get the real shape of "None".

I try two methods when I build model:

First, x.get_shape() and get shape as [Dimension(None), Dimension(128)]

Second, x.shape and get shape as [Dimension(None), Dimension(128)]

And what I want is the real shape, for example, when the batch size is 100 in this round, I would like to get [Dimension(100), Dimension(128)].

How do I get the dynamic input shape?

Answer 1

I believe tf.shape is what you are looking for.

tf.shape(x) can get the shape while session is running.

The full example is below:

import tensorflow as tf
a = tf.ones([3,4])
b = tf.shape(a)
sess=tf.Session()
print(b.eval(session=sess))

You can also use b to init new variables.